16x16 and 8x8c x86 SIMD intra pred functions for VP8 and H.264

Originally committed as revision 23783 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/h264pred.c

@@ -1299,4 +1299,5 @@ void ff_h264_pred_init(H264PredContext *h, int codec_id){
     h->pred16x16_add[ HOR_PRED8x8]= pred16x16_horizontal_add_c;
 
     if (ARCH_ARM) ff_h264_pred_init_arm(h, codec_id);
+    if (ARCH_X86) ff_h264_pred_init_x86(h, codec_id);
 }

libavcodec/h264pred.h

@@ -87,5 +87,6 @@ typedef struct H264PredContext{
 
 void ff_h264_pred_init(H264PredContext *h, int codec_id);
 void ff_h264_pred_init_arm(H264PredContext *h, int codec_id);
+void ff_h264_pred_init_x86(H264PredContext *h, int codec_id);
 
 #endif /* AVCODEC_H264PRED_H */

libavcodec/x86/Makefile

@@ -8,6 +8,7 @@ YASM-OBJS-$(CONFIG_FFT)                += x86/fft_mmx.o                 \
                                           $(YASM-OBJS-FFT-yes)
 YASM-OBJS-$(CONFIG_GPL)                += x86/h264_deblock_sse2.o       \
                                           x86/h264_idct_sse2.o          \
+                                          x86/h264_intrapred.o          \
 
 MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
 MMX-OBJS-$(CONFIG_MP1FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o

libavcodec/x86/dsputil_mmx.c

@@ -62,7 +62,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
 
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3  ) = 0x0303030303030303ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;

libavcodec/x86/dsputil_mmx.h

@@ -48,7 +48,7 @@ extern const uint64_t ff_pw_128;
 extern const uint64_t ff_pw_255;
 
 extern const uint64_t ff_pb_1;
-extern const uint64_t ff_pb_3;
+extern const xmm_reg  ff_pb_3;
 extern const uint64_t ff_pb_7;
 extern const uint64_t ff_pb_1F;
 extern const uint64_t ff_pb_3F;

libavcodec/x86/h264dsp_mmx.c

@@ -19,6 +19,7 @@
  */
 
 #include "dsputil_mmx.h"
+#include "libavcodec/h264pred.h"
 
 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_7_3  ) = 0x0307030703070307ULL;
@@ -2322,3 +2323,77 @@ H264_WEIGHT( 4, 8)
 H264_WEIGHT( 4, 4)
 H264_WEIGHT( 4, 2)
 
+void ff_pred16x16_vertical_mmx     (uint8_t *src, int stride);
+void ff_pred16x16_vertical_sse     (uint8_t *src, int stride);
+void ff_pred16x16_horizontal_mmx   (uint8_t *src, int stride);
+void ff_pred16x16_horizontal_mmxext(uint8_t *src, int stride);
+void ff_pred16x16_horizontal_ssse3 (uint8_t *src, int stride);
+void ff_pred16x16_dc_mmx           (uint8_t *src, int stride);
+void ff_pred16x16_dc_mmxext        (uint8_t *src, int stride);
+void ff_pred16x16_dc_sse           (uint8_t *src, int stride);
+void ff_pred16x16_dc_sse2          (uint8_t *src, int stride);
+void ff_pred16x16_dc_ssse3         (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_mmx       (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_mmxext    (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_sse2      (uint8_t *src, int stride);
+void ff_pred8x8_dc_rv40_mmx        (uint8_t *src, int stride);
+void ff_pred8x8_dc_rv40_mmxext     (uint8_t *src, int stride);
+void ff_pred8x8_vertical_mmx       (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_mmx     (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_mmxext  (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_ssse3   (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_mmx         (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_mmxext      (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_sse2        (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_ssse3      (uint8_t *src, int stride);
+
+void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
+{
+#if HAVE_YASM
+    if (mm_flags & FF_MM_MMX) {
+        h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx;
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmx;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_mmx;
+        h->pred8x8  [VERT_PRED8x8] = ff_pred8x8_vertical_mmx;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmx;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmx;
+            h->pred8x8  [DC_PRED8x8   ] = ff_pred8x8_dc_rv40_mmx;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmx;
+        }
+    }
+
+    if (mm_flags & FF_MM_MMXEXT) {
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmxext;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_mmxext;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmxext;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmxext;
+            h->pred8x8  [DC_PRED8x8   ] = ff_pred8x8_dc_rv40_mmxext;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmxext;
+        }
+    }
+
+    if (mm_flags & FF_MM_SSE) {
+        h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_sse;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_sse;
+    }
+
+    if (mm_flags & FF_MM_SSE2) {
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_sse2;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_sse2;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_sse2;
+        }
+    }
+
+    if (mm_flags & FF_MM_SSSE3) {
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_ssse3;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_ssse3;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_ssse3;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_ssse3;
+        }
+    }
+#endif
+}